Utility-Oriented Gradual Itemsets Mining Using High Utility Itemsets Mining

Abstract

Gradual itemsets capture frequent covariations of attributes of the form “more/less A, more/less B” from a quantitative database. These patterns have gained considerable interest during these years and have been applied in several domains. Various algorithms have been proposed to extract those itemsets efficiently. However, an inherent limitation of the proposed algorithms is that they only evaluate items in terms of increase and decrease. Therefore, all the covariations of items have an equal importance/signification in evaluating the frequency of a gradual itemset. Those algorithms are not appropriate for certain real-world applications where strong covariations which are scarce may be useful. This paper proposes a solution to cope with this limitation with the task of high utility gradual itemsets mining, whose goal is to extract covariations of attributes which generate a high profit for the user. Two algorithms are proposed to mine these patterns efficiently called HUGI (High Utility Gradual Itemsets mining), and HUGI-Merging, which extracts these patterns from both a negative and positive quantitative data separately and merges the obtained results. Experimental results show that the proposed algorithms are efficient and can filter many gradual itemsets to focus only on desired high-utility gradual itemsets.